AI takes root across agencies
- By Matt Leonard
- Mar 08, 2018
When House members and agency executives gathered on Capitol Hill March 7 to talk about how artificial intelligence could benefit the public sector, legislators focused their comments on the president's budget, which cuts research and development funding.
“Outside of the Department of Defense, the president’s budget proposes an overall cut to research and development of 21.2 percent,” Rep. Robin Kelly (D-Ill.) said in her opening statement at the hearing of the House Oversight and Government Reform's Subcommittee on Information Technology. Rep. Gerry Connolly (D-Va.) agreed the cuts were worrying: “I tremble at what we are cutting.”
The subcommittee did hear from witnesses who described their agencies' AI projects.
The General Services Administration, for instance, is exploring how robotic process automation can offload compliance monitoring tasks from employees, Keith Nakasone, the deputy assistant commissioner of Information Technology Category Acquisition Management within GSA, explained in his testimony.
GSA's Solicitation Review Tool predicts regulatory compliance by using "natural language processing, text mining, and machine learning algorithms to automatically predict whether federal solicitations posted on fbo.gov are compliant with Section 508 of the Rehabilitation Act," Nakasone testified. If filings are found to be out of compliance, the tool alerts responsible parties so corrections can be made. “Through independent review, the predictions have an accuracy rate of 95 percent,” he added.
The tool is expected to go into production on cloud.gov this spring.
GSA also has built a team focused on robotic process automation that will provide governance for these new technologies, including common architectures.
AI has shown potential to help agencies improve cybersecurity by predicting malware evolution, according to Douglas Maughan, the director of the Cybersecurity Division at the Homeland Security Advanced Research Project Agency.
Additionally, the DHS Science and Technology Directorate built "a machine learning-based policy engine capable of blocking more than 120,000 calls a month, including robo calls,” he told lawmakers. “This same technology can be used defend 911 centers against life threatening distributed denial of service attacks” that overwhelm emergency call centers with fraudulent calls.
At the Defense Advanced Research Projects Agency, the Explainable Artificial Intelligence program aims to create machine-learning systems that can better describe how they came to their conclusions. This will help alleviate some of the trust issues associated with AI technologies, said John O. Everett, the deputy director of the DARPA Information Innovation Office
Future research that will be vital, he said, will be providing computers with common sense, which will require helping them better understand how intonation and facial expressions provide context for speech.
To end the hearing, Rep. Will Hurd (R-Texas) asked panelists what the equivalent of going to the moon is for AI.
Today we have narrow AI, which is image recognition and speech recognition, but the next big step will be general AI, said James F. Kurose, the assistant director of Computer Science and Information Science and Engineering at the National Science Foundation.
“Look at, for instance, what an 18-month-old child can do and how the child can transfer learning from one environment to another, how a child can understand intent and meaning,” he said. “That’s really the grand challenge. General AI still remains a very grand challenge.”
Matt Leonard is a reporter/producer at GCN.
Before joining GCN, Leonard worked as a local reporter for The Smithfield Times in southeastern Virginia. In his time there he wrote about town council meetings, local crime and what to do if a beaver dam floods your back yard. Over the last few years, he has spent time at The Commonwealth Times, The Denver Post and WTVR-CBS 6. He is a graduate of Virginia Commonwealth University, where he received the faculty award for print and online journalism.
Leonard can be contacted at firstname.lastname@example.org or follow him on Twitter @Matt_Lnrd.
Click here for previous articles by Leonard.